1 /*
2 * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "compiler/compiler_globals.hpp"
26 #include "interp_masm_x86.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "logging/log.hpp"
30 #include "oops/arrayOop.hpp"
31 #include "oops/markWord.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/method.hpp"
34 #include "oops/resolvedFieldEntry.hpp"
35 #include "oops/resolvedIndyEntry.hpp"
36 #include "oops/resolvedMethodEntry.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "runtime/basicLock.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/javaThread.hpp"
42 #include "runtime/safepointMechanism.hpp"
43 #include "runtime/sharedRuntime.hpp"
44 #include "utilities/powerOfTwo.hpp"
45
46 // Implementation of InterpreterMacroAssembler
47
48 void InterpreterMacroAssembler::jump_to_entry(address entry) {
49 assert(entry, "Entry must have been generated by now");
50 jump(RuntimeAddress(entry));
51 }
52
53 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
54 Label update, next, none;
55
56 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
57
58 interp_verify_oop(obj, atos);
59
60 testptr(obj, obj);
61 jccb(Assembler::notZero, update);
62 testptr(mdo_addr, TypeEntries::null_seen);
63 jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
64 // atomic update to prevent overwriting Klass* with 0
65 lock();
66 orptr(mdo_addr, TypeEntries::null_seen);
67 jmpb(next);
68
69 bind(update);
70 load_klass(obj, obj, rscratch1);
71 mov(rscratch1, obj);
72
73 xorptr(obj, mdo_addr);
74 testptr(obj, TypeEntries::type_klass_mask);
75 jccb(Assembler::zero, next); // klass seen before, nothing to
76 // do. The unknown bit may have been
77 // set already but no need to check.
78
79 testptr(obj, TypeEntries::type_unknown);
80 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
81
82 cmpptr(mdo_addr, 0);
83 jccb(Assembler::equal, none);
84 cmpptr(mdo_addr, TypeEntries::null_seen);
85 jccb(Assembler::equal, none);
86
87 // There is a chance that the checks above (re-reading profiling
88 // data from memory) fail if another thread has just set the
89 // profiling to this obj's klass
90 mov(obj, rscratch1);
91 xorptr(obj, mdo_addr);
92 testptr(obj, TypeEntries::type_klass_mask);
93 jccb(Assembler::zero, next);
94
95 // different than before. Cannot keep accurate profile.
96 orptr(mdo_addr, TypeEntries::type_unknown);
97 jmpb(next);
98
99 bind(none);
100 // first time here. Set profile type.
101 movptr(mdo_addr, obj);
102 #ifdef ASSERT
103 andptr(obj, TypeEntries::type_klass_mask);
104 verify_klass_ptr(obj);
105 #endif
106
107 bind(next);
108 }
109
110 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
111 if (!ProfileInterpreter) {
112 return;
113 }
114
115 if (MethodData::profile_arguments() || MethodData::profile_return()) {
116 Label profile_continue;
117
118 test_method_data_pointer(mdp, profile_continue);
119
120 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
121
122 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
123 jcc(Assembler::notEqual, profile_continue);
124
125 if (MethodData::profile_arguments()) {
126 Label done;
127 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
128 addptr(mdp, off_to_args);
129
130 for (int i = 0; i < TypeProfileArgsLimit; i++) {
131 if (i > 0 || MethodData::profile_return()) {
132 // If return value type is profiled we may have no argument to profile
133 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
134 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
135 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
136 jcc(Assembler::less, done);
137 }
138 movptr(tmp, Address(callee, Method::const_offset()));
139 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
140 // stack offset o (zero based) from the start of the argument
141 // list, for n arguments translates into offset n - o - 1 from
142 // the end of the argument list
143 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
144 subl(tmp, 1);
145 Address arg_addr = argument_address(tmp);
146 movptr(tmp, arg_addr);
147
148 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
149 profile_obj_type(tmp, mdo_arg_addr);
150
151 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
152 addptr(mdp, to_add);
153 off_to_args += to_add;
154 }
155
156 if (MethodData::profile_return()) {
157 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
158 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
159 }
160
161 bind(done);
162
163 if (MethodData::profile_return()) {
164 // We're right after the type profile for the last
165 // argument. tmp is the number of cells left in the
166 // CallTypeData/VirtualCallTypeData to reach its end. Non null
167 // if there's a return to profile.
168 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
169 shll(tmp, log2i_exact((int)DataLayout::cell_size));
170 addptr(mdp, tmp);
171 }
172 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
173 } else {
174 assert(MethodData::profile_return(), "either profile call args or call ret");
175 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
176 }
177
178 // mdp points right after the end of the
179 // CallTypeData/VirtualCallTypeData, right after the cells for the
180 // return value type if there's one
181
182 bind(profile_continue);
183 }
184 }
185
186 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
187 assert_different_registers(mdp, ret, tmp, _bcp_register);
188 if (ProfileInterpreter && MethodData::profile_return()) {
189 Label profile_continue;
190
191 test_method_data_pointer(mdp, profile_continue);
192
193 if (MethodData::profile_return_jsr292_only()) {
194 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
195
196 // If we don't profile all invoke bytecodes we must make sure
197 // it's a bytecode we indeed profile. We can't go back to the
198 // beginning of the ProfileData we intend to update to check its
199 // type because we're right after it and we don't known its
200 // length
201 Label do_profile;
202 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
203 jcc(Assembler::equal, do_profile);
204 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
205 jcc(Assembler::equal, do_profile);
206 get_method(tmp);
207 cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
208 jcc(Assembler::notEqual, profile_continue);
209
210 bind(do_profile);
211 }
212
213 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
214 mov(tmp, ret);
215 profile_obj_type(tmp, mdo_ret_addr);
216
217 bind(profile_continue);
218 }
219 }
220
221 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
222 if (ProfileInterpreter && MethodData::profile_parameters()) {
223 Label profile_continue;
224
225 test_method_data_pointer(mdp, profile_continue);
226
227 // Load the offset of the area within the MDO used for
228 // parameters. If it's negative we're not profiling any parameters
229 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
230 testl(tmp1, tmp1);
231 jcc(Assembler::negative, profile_continue);
232
233 // Compute a pointer to the area for parameters from the offset
234 // and move the pointer to the slot for the last
235 // parameters. Collect profiling from last parameter down.
236 // mdo start + parameters offset + array length - 1
237 addptr(mdp, tmp1);
238 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
239 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
240
241 Label loop;
242 bind(loop);
243
244 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
245 int type_base = in_bytes(ParametersTypeData::type_offset(0));
246 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
247 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
248 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
249
250 // load offset on the stack from the slot for this parameter
251 movptr(tmp2, arg_off);
252 negptr(tmp2);
253 // read the parameter from the local area
254 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
255
256 // profile the parameter
257 profile_obj_type(tmp2, arg_type);
258
259 // go to next parameter
260 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
261 jcc(Assembler::positive, loop);
262
263 bind(profile_continue);
264 }
265 }
266
267 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
268 int number_of_arguments) {
269 // interpreter specific
270 //
271 // Note: No need to save/restore bcp & locals registers
272 // since these are callee saved registers and no blocking/
273 // GC can happen in leaf calls.
274 // Further Note: DO NOT save/restore bcp/locals. If a caller has
275 // already saved them so that it can use rsi/rdi as temporaries
276 // then a save/restore here will DESTROY the copy the caller
277 // saved! There used to be a save_bcp() that only happened in
278 // the ASSERT path (no restore_bcp). Which caused bizarre failures
279 // when jvm built with ASSERTs.
280 #ifdef ASSERT
281 {
282 Label L;
283 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
284 jcc(Assembler::equal, L);
285 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
286 " last_sp != null");
287 bind(L);
288 }
289 #endif
290 // super call
291 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
292 // interpreter specific
293 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
294 // but since they may not have been saved (and we don't want to
295 // save them here (see note above) the assert is invalid.
296 }
297
298 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
299 Register last_java_sp,
300 address entry_point,
301 int number_of_arguments,
302 bool check_exceptions) {
303 // interpreter specific
304 //
305 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
306 // really make a difference for these runtime calls, since they are
307 // slow anyway. Btw., bcp must be saved/restored since it may change
308 // due to GC.
309 save_bcp();
310 #ifdef ASSERT
311 {
312 Label L;
313 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
314 jcc(Assembler::equal, L);
315 stop("InterpreterMacroAssembler::call_VM_base:"
316 " last_sp isn't null");
317 bind(L);
318 }
319 #endif /* ASSERT */
320 // super call
321 MacroAssembler::call_VM_base(oop_result, last_java_sp,
322 entry_point, number_of_arguments,
323 check_exceptions);
324 // interpreter specific
325 restore_bcp();
326 restore_locals();
327 }
328
329 void InterpreterMacroAssembler::call_VM_preemptable_helper(Register oop_result,
330 address entry_point,
331 int number_of_arguments,
332 bool check_exceptions) {
333 Label resume_pc, not_preempted;
334
335 #ifdef ASSERT
336 {
337 Label L;
338 cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
339 jcc(Assembler::equal, L);
340 stop("Should not have alternate return address set");
341 bind(L);
342 }
343 #endif /* ASSERT */
344
345 // Force freeze slow path.
346 push_cont_fastpath();
347
348 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
349 // Note: call_VM_helper requires last_Java_pc for anchor to be at the top of the stack.
350 lea(rscratch1, resume_pc);
351 push(rscratch1);
352 MacroAssembler::call_VM_helper(oop_result, entry_point, number_of_arguments, check_exceptions);
353 pop(rscratch1);
354
355 pop_cont_fastpath();
356
357 // Check if preempted.
358 movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
359 cmpptr(rscratch1, NULL_WORD);
360 jccb(Assembler::zero, not_preempted);
361 movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
362 jmp(rscratch1);
363
364 // In case of preemption, this is where we will resume once we finally acquire the monitor.
365 bind(resume_pc);
366 restore_after_resume(false /* is_native */);
367
368 if (check_exceptions) {
369 // check for pending exceptions (java_thread is set upon return)
370 cmpptr(Address(r15_thread, Thread::pending_exception_offset()), NULL_WORD);
371 Label ok;
372 jcc(Assembler::equal, ok);
373 // Exception stub expects return pc to be at top of stack. We only need
374 // it to check Interpreter::contains(return_address) so anything will do.
375 lea(rscratch1, resume_pc);
376 push(rscratch1);
377 jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
378 bind(ok);
379 }
380
381 // get oop result if there is one and reset the value in the thread
382 if (oop_result->is_valid()) {
383 get_vm_result_oop(oop_result);
384 }
385
386 bind(not_preempted);
387 }
388
389 static void pass_arg1(MacroAssembler* masm, Register arg) {
390 if (c_rarg1 != arg ) {
391 masm->mov(c_rarg1, arg);
392 }
393 }
394
395 static void pass_arg2(MacroAssembler* masm, Register arg) {
396 if (c_rarg2 != arg ) {
397 masm->mov(c_rarg2, arg);
398 }
399 }
400
401 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
402 address entry_point,
403 Register arg_1,
404 bool check_exceptions) {
405 pass_arg1(this, arg_1);
406 call_VM_preemptable_helper(oop_result, entry_point, 1, check_exceptions);
407 }
408
409 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
410 address entry_point,
411 Register arg_1,
412 Register arg_2,
413 bool check_exceptions) {
414 LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
415 pass_arg2(this, arg_2);
416 pass_arg1(this, arg_1);
417 call_VM_preemptable_helper(oop_result, entry_point, 2, check_exceptions);
418 }
419
420 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
421 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
422 call(rscratch1);
423 if (is_native) {
424 // On resume we need to set up stack as expected.
425 push(dtos);
426 push(ltos);
427 }
428 }
429
430 void InterpreterMacroAssembler::check_and_handle_popframe() {
431 if (JvmtiExport::can_pop_frame()) {
432 Label L;
433 // Initiate popframe handling only if it is not already being
434 // processed. If the flag has the popframe_processing bit set, it
435 // means that this code is called *during* popframe handling - we
436 // don't want to reenter.
437 // This method is only called just after the call into the vm in
438 // call_VM_base, so the arg registers are available.
439 Register pop_cond = c_rarg0;
440 movl(pop_cond, Address(r15_thread, JavaThread::popframe_condition_offset()));
441 testl(pop_cond, JavaThread::popframe_pending_bit);
442 jcc(Assembler::zero, L);
443 testl(pop_cond, JavaThread::popframe_processing_bit);
444 jcc(Assembler::notZero, L);
445 // Call Interpreter::remove_activation_preserving_args_entry() to get the
446 // address of the same-named entrypoint in the generated interpreter code.
447 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
448 jmp(rax);
449 bind(L);
450 }
451 }
452
453 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
454 movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
455 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
456 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
457 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
458
459 switch (state) {
460 case atos: movptr(rax, oop_addr);
461 movptr(oop_addr, NULL_WORD);
462 interp_verify_oop(rax, state); break;
463 case ltos: movptr(rax, val_addr); break;
464 case btos: // fall through
465 case ztos: // fall through
466 case ctos: // fall through
467 case stos: // fall through
468 case itos: movl(rax, val_addr); break;
469 case ftos: movflt(xmm0, val_addr); break;
470 case dtos: movdbl(xmm0, val_addr); break;
471 case vtos: /* nothing to do */ break;
472 default : ShouldNotReachHere();
473 }
474
475 // Clean up tos value in the thread object
476 movl(tos_addr, ilgl);
477 movptr(val_addr, NULL_WORD);
478 }
479
480
481 void InterpreterMacroAssembler::check_and_handle_earlyret() {
482 if (JvmtiExport::can_force_early_return()) {
483 Label L;
484 Register tmp = c_rarg0;
485 Register rthread = r15_thread;
486
487 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
488 testptr(tmp, tmp);
489 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
490
491 // Initiate earlyret handling only if it is not already being processed.
492 // If the flag has the earlyret_processing bit set, it means that this code
493 // is called *during* earlyret handling - we don't want to reenter.
494 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
495 cmpl(tmp, JvmtiThreadState::earlyret_pending);
496 jcc(Assembler::notEqual, L);
497
498 // Call Interpreter::remove_activation_early_entry() to get the address of the
499 // same-named entrypoint in the generated interpreter code.
500 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
501 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
502 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
503 jmp(rax);
504 bind(L);
505 }
506 }
507
508 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
509 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
510 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
511 bswapl(reg);
512 shrl(reg, 16);
513 }
514
515 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
516 int bcp_offset,
517 size_t index_size) {
518 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
519 if (index_size == sizeof(u2)) {
520 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
521 } else if (index_size == sizeof(u4)) {
522 movl(index, Address(_bcp_register, bcp_offset));
523 } else if (index_size == sizeof(u1)) {
524 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
525 } else {
526 ShouldNotReachHere();
527 }
528 }
529
530 // Load object from cpool->resolved_references(index)
531 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
532 Register index,
533 Register tmp) {
534 assert_different_registers(result, index);
535
536 get_constant_pool(result);
537 // load pointer for resolved_references[] objArray
538 movptr(result, Address(result, ConstantPool::cache_offset()));
539 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
540 resolve_oop_handle(result, tmp);
541 load_heap_oop(result, Address(result, index,
542 UseCompressedOops ? Address::times_4 : Address::times_ptr,
543 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
544 }
545
546 // load cpool->resolved_klass_at(index)
547 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
548 Register cpool,
549 Register index) {
550 assert_different_registers(cpool, index);
551
552 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
553 Register resolved_klasses = cpool;
554 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
555 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
556 }
557
558 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
559 // subtype of super_klass.
560 //
561 // Args:
562 // rax: superklass
563 // Rsub_klass: subklass
564 //
565 // Kills:
566 // rcx, rdi
567 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
568 Label& ok_is_subtype) {
569 assert(Rsub_klass != rax, "rax holds superklass");
570 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
571 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
572 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
573 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
574
575 // Profile the not-null value's klass.
576 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
577
578 // Do the check.
579 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
580 }
581
582
583 // Java Expression Stack
584
585 void InterpreterMacroAssembler::pop_ptr(Register r) {
586 pop(r);
587 }
588
589 void InterpreterMacroAssembler::push_ptr(Register r) {
590 push(r);
591 }
592
593 void InterpreterMacroAssembler::push_i(Register r) {
594 push(r);
595 }
596
597 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
598 push(r);
599 }
600
601 void InterpreterMacroAssembler::push_f(XMMRegister r) {
602 subptr(rsp, wordSize);
603 movflt(Address(rsp, 0), r);
604 }
605
606 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
607 movflt(r, Address(rsp, 0));
608 addptr(rsp, wordSize);
609 }
610
611 void InterpreterMacroAssembler::push_d(XMMRegister r) {
612 subptr(rsp, 2 * wordSize);
613 movdbl(Address(rsp, 0), r);
614 }
615
616 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
617 movdbl(r, Address(rsp, 0));
618 addptr(rsp, 2 * Interpreter::stackElementSize);
619 }
620
621 void InterpreterMacroAssembler::pop_i(Register r) {
622 // XXX can't use pop currently, upper half non clean
623 movl(r, Address(rsp, 0));
624 addptr(rsp, wordSize);
625 }
626
627 void InterpreterMacroAssembler::pop_l(Register r) {
628 movq(r, Address(rsp, 0));
629 addptr(rsp, 2 * Interpreter::stackElementSize);
630 }
631
632 void InterpreterMacroAssembler::push_l(Register r) {
633 subptr(rsp, 2 * wordSize);
634 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
635 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
636 }
637
638 void InterpreterMacroAssembler::pop(TosState state) {
639 switch (state) {
640 case atos: pop_ptr(); break;
641 case btos:
642 case ztos:
643 case ctos:
644 case stos:
645 case itos: pop_i(); break;
646 case ltos: pop_l(); break;
647 case ftos: pop_f(xmm0); break;
648 case dtos: pop_d(xmm0); break;
649 case vtos: /* nothing to do */ break;
650 default: ShouldNotReachHere();
651 }
652 interp_verify_oop(rax, state);
653 }
654
655 void InterpreterMacroAssembler::push(TosState state) {
656 interp_verify_oop(rax, state);
657 switch (state) {
658 case atos: push_ptr(); break;
659 case btos:
660 case ztos:
661 case ctos:
662 case stos:
663 case itos: push_i(); break;
664 case ltos: push_l(); break;
665 case ftos: push_f(xmm0); break;
666 case dtos: push_d(xmm0); break;
667 case vtos: /* nothing to do */ break;
668 default : ShouldNotReachHere();
669 }
670 }
671
672 // Helpers for swap and dup
673 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
674 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
675 }
676
677 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
678 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
679 }
680
681
682 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
683 // set sender sp
684 lea(_bcp_register, Address(rsp, wordSize));
685 // record last_sp
686 mov(rcx, _bcp_register);
687 subptr(rcx, rbp);
688 sarptr(rcx, LogBytesPerWord);
689 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
690 }
691
692
693 // Jump to from_interpreted entry of a call unless single stepping is possible
694 // in this thread in which case we must call the i2i entry
695 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
696 prepare_to_jump_from_interpreted();
697
698 if (JvmtiExport::can_post_interpreter_events()) {
699 Label run_compiled_code;
700 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
701 // compiled code in threads for which the event is enabled. Check here for
702 // interp_only_mode if these events CAN be enabled.
703 // interp_only is an int, on little endian it is sufficient to test the byte only
704 // Is a cmpl faster?
705 cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
706 jccb(Assembler::zero, run_compiled_code);
707 jmp(Address(method, Method::interpreter_entry_offset()));
708 bind(run_compiled_code);
709 }
710
711 jmp(Address(method, Method::from_interpreted_offset()));
712 }
713
714 // The following two routines provide a hook so that an implementation
715 // can schedule the dispatch in two parts. x86 does not do this.
716 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
717 // Nothing x86 specific to be done here
718 }
719
720 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
721 dispatch_next(state, step);
722 }
723
724 void InterpreterMacroAssembler::dispatch_base(TosState state,
725 address* table,
726 bool verifyoop,
727 bool generate_poll) {
728 if (VerifyActivationFrameSize) {
729 Label L;
730 mov(rcx, rbp);
731 subptr(rcx, rsp);
732 int32_t min_frame_size =
733 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
734 wordSize;
735 cmpptr(rcx, min_frame_size);
736 jcc(Assembler::greaterEqual, L);
737 stop("broken stack frame");
738 bind(L);
739 }
740 if (verifyoop) {
741 interp_verify_oop(rax, state);
742 }
743
744 address* const safepoint_table = Interpreter::safept_table(state);
745 Label no_safepoint, dispatch;
746 if (table != safepoint_table && generate_poll) {
747 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
748 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
749
750 jccb(Assembler::zero, no_safepoint);
751 lea(rscratch1, ExternalAddress((address)safepoint_table));
752 jmpb(dispatch);
753 }
754
755 bind(no_safepoint);
756 lea(rscratch1, ExternalAddress((address)table));
757 bind(dispatch);
758 jmp(Address(rscratch1, rbx, Address::times_8));
759 }
760
761 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
762 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
763 }
764
765 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
766 dispatch_base(state, Interpreter::normal_table(state));
767 }
768
769 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
770 dispatch_base(state, Interpreter::normal_table(state), false);
771 }
772
773
774 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
775 // load next bytecode (load before advancing _bcp_register to prevent AGI)
776 load_unsigned_byte(rbx, Address(_bcp_register, step));
777 // advance _bcp_register
778 increment(_bcp_register, step);
779 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
780 }
781
782 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
783 // load current bytecode
784 load_unsigned_byte(rbx, Address(_bcp_register, 0));
785 dispatch_base(state, table);
786 }
787
788 void InterpreterMacroAssembler::narrow(Register result) {
789
790 // Get method->_constMethod->_result_type
791 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
792 movptr(rcx, Address(rcx, Method::const_offset()));
793 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
794
795 Label done, notBool, notByte, notChar;
796
797 // common case first
798 cmpl(rcx, T_INT);
799 jcc(Assembler::equal, done);
800
801 // mask integer result to narrower return type.
802 cmpl(rcx, T_BOOLEAN);
803 jcc(Assembler::notEqual, notBool);
804 andl(result, 0x1);
805 jmp(done);
806
807 bind(notBool);
808 cmpl(rcx, T_BYTE);
809 jcc(Assembler::notEqual, notByte);
810 movsbl(result, result);
811 jmp(done);
812
813 bind(notByte);
814 cmpl(rcx, T_CHAR);
815 jcc(Assembler::notEqual, notChar);
816 movzwl(result, result);
817 jmp(done);
818
819 bind(notChar);
820 // cmpl(rcx, T_SHORT); // all that's left
821 // jcc(Assembler::notEqual, done);
822 movswl(result, result);
823
824 // Nothing to do for T_INT
825 bind(done);
826 }
827
828 // remove activation
829 //
830 // Unlock the receiver if this is a synchronized method.
831 // Unlock any Java monitors from synchronized blocks.
832 // Apply stack watermark barrier.
833 // Notify JVMTI.
834 // Remove the activation from the stack.
835 //
836 // If there are locked Java monitors
837 // If throw_monitor_exception
838 // throws IllegalMonitorStateException
839 // Else if install_monitor_exception
840 // installs IllegalMonitorStateException
841 // Else
842 // no error processing
843 void InterpreterMacroAssembler::remove_activation(TosState state,
844 Register ret_addr,
845 bool throw_monitor_exception,
846 bool install_monitor_exception,
847 bool notify_jvmdi) {
848 // Note: Registers rdx xmm0 may be in use for the
849 // result check if synchronized method
850 Label unlocked, unlock, no_unlock;
851
852 const Register rthread = r15_thread;
853 const Register robj = c_rarg1;
854 const Register rmon = c_rarg1;
855
856 // get the value of _do_not_unlock_if_synchronized into rdx
857 const Address do_not_unlock_if_synchronized(rthread,
858 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
859 movbool(rbx, do_not_unlock_if_synchronized);
860 movbool(do_not_unlock_if_synchronized, false); // reset the flag
861
862 // get method access flags
863 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
864 load_unsigned_short(rcx, Address(rcx, Method::access_flags_offset()));
865 testl(rcx, JVM_ACC_SYNCHRONIZED);
866 jcc(Assembler::zero, unlocked);
867
868 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
869 // is set.
870 testbool(rbx);
871 jcc(Assembler::notZero, no_unlock);
872
873 // unlock monitor
874 push(state); // save result
875
876 // BasicObjectLock will be first in list, since this is a
877 // synchronized method. However, need to check that the object has
878 // not been unlocked by an explicit monitorexit bytecode.
879 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
880 wordSize - (int) sizeof(BasicObjectLock));
881 // We use c_rarg1/rdx so that if we go slow path it will be the correct
882 // register for unlock_object to pass to VM directly
883 lea(robj, monitor); // address of first monitor
884
885 movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
886 testptr(rax, rax);
887 jcc(Assembler::notZero, unlock);
888
889 pop(state);
890 if (throw_monitor_exception) {
891 // Entry already unlocked, need to throw exception
892 call_VM(noreg, CAST_FROM_FN_PTR(address,
893 InterpreterRuntime::throw_illegal_monitor_state_exception));
894 should_not_reach_here();
895 } else {
896 // Monitor already unlocked during a stack unroll. If requested,
897 // install an illegal_monitor_state_exception. Continue with
898 // stack unrolling.
899 if (install_monitor_exception) {
900 call_VM(noreg, CAST_FROM_FN_PTR(address,
901 InterpreterRuntime::new_illegal_monitor_state_exception));
902 }
903 jmp(unlocked);
904 }
905
906 bind(unlock);
907 unlock_object(robj);
908 pop(state);
909
910 // Check that for block-structured locking (i.e., that all locked
911 // objects has been unlocked)
912 bind(unlocked);
913
914 // rax, rdx: Might contain return value
915
916 // Check that all monitors are unlocked
917 {
918 Label loop, exception, entry, restart;
919 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
920 const Address monitor_block_top(
921 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
922 const Address monitor_block_bot(
923 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
924
925 bind(restart);
926 // We use c_rarg1 so that if we go slow path it will be the correct
927 // register for unlock_object to pass to VM directly
928 movptr(rmon, monitor_block_top); // derelativize pointer
929 lea(rmon, Address(rbp, rmon, Address::times_ptr));
930 // c_rarg1 points to current entry, starting with top-most entry
931
932 lea(rbx, monitor_block_bot); // points to word before bottom of
933 // monitor block
934 jmp(entry);
935
936 // Entry already locked, need to throw exception
937 bind(exception);
938
939 if (throw_monitor_exception) {
940 // Throw exception
941 MacroAssembler::call_VM(noreg,
942 CAST_FROM_FN_PTR(address, InterpreterRuntime::
943 throw_illegal_monitor_state_exception));
944 should_not_reach_here();
945 } else {
946 // Stack unrolling. Unlock object and install illegal_monitor_exception.
947 // Unlock does not block, so don't have to worry about the frame.
948 // We don't have to preserve c_rarg1 since we are going to throw an exception.
949
950 push(state);
951 mov(robj, rmon); // nop if robj and rmon are the same
952 unlock_object(robj);
953 pop(state);
954
955 if (install_monitor_exception) {
956 call_VM(noreg, CAST_FROM_FN_PTR(address,
957 InterpreterRuntime::
958 new_illegal_monitor_state_exception));
959 }
960
961 jmp(restart);
962 }
963
964 bind(loop);
965 // check if current entry is used
966 cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
967 jcc(Assembler::notEqual, exception);
968
969 addptr(rmon, entry_size); // otherwise advance to next entry
970 bind(entry);
971 cmpptr(rmon, rbx); // check if bottom reached
972 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
973 }
974
975 bind(no_unlock);
976
977 JFR_ONLY(enter_jfr_critical_section();)
978
979 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
980 // that would normally not be safe to use. Such bad returns into unsafe territory of
981 // the stack, will call InterpreterRuntime::at_unwind.
982 Label slow_path;
983 Label fast_path;
984 safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */);
985 jmp(fast_path);
986 bind(slow_path);
987 push(state);
988 set_last_Java_frame(noreg, rbp, (address)pc(), rscratch1);
989 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), r15_thread);
990 reset_last_Java_frame(true);
991 pop(state);
992 bind(fast_path);
993
994 // JVMTI support. Make sure the safepoint poll test is issued prior.
995 if (notify_jvmdi) {
996 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
997 } else {
998 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
999 }
1000
1001 // remove activation
1002 // get sender sp
1003 movptr(rbx,
1004 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1005 if (StackReservedPages > 0) {
1006 // testing if reserved zone needs to be re-enabled
1007 Register rthread = r15_thread;
1008 Label no_reserved_zone_enabling;
1009
1010 // check if already enabled - if so no re-enabling needed
1011 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
1012 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1013 jcc(Assembler::equal, no_reserved_zone_enabling);
1014
1015 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1016 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1017
1018 JFR_ONLY(leave_jfr_critical_section();)
1019
1020 call_VM_leaf(
1021 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1022 call_VM(noreg, CAST_FROM_FN_PTR(address,
1023 InterpreterRuntime::throw_delayed_StackOverflowError));
1024 should_not_reach_here();
1025
1026 bind(no_reserved_zone_enabling);
1027 }
1028
1029 leave(); // remove frame anchor
1030
1031 JFR_ONLY(leave_jfr_critical_section();)
1032
1033 pop(ret_addr); // get return address
1034 mov(rsp, rbx); // set sp to sender sp
1035 pop_cont_fastpath();
1036
1037 }
1038
1039 #if INCLUDE_JFR
1040 void InterpreterMacroAssembler::enter_jfr_critical_section() {
1041 const Address sampling_critical_section(r15_thread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
1042 movbool(sampling_critical_section, true);
1043 }
1044
1045 void InterpreterMacroAssembler::leave_jfr_critical_section() {
1046 const Address sampling_critical_section(r15_thread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
1047 movbool(sampling_critical_section, false);
1048 }
1049 #endif // INCLUDE_JFR
1050
1051 void InterpreterMacroAssembler::get_method_counters(Register method,
1052 Register mcs, Label& skip) {
1053 Label has_counters;
1054 movptr(mcs, Address(method, Method::method_counters_offset()));
1055 testptr(mcs, mcs);
1056 jcc(Assembler::notZero, has_counters);
1057 call_VM(noreg, CAST_FROM_FN_PTR(address,
1058 InterpreterRuntime::build_method_counters), method);
1059 movptr(mcs, Address(method,Method::method_counters_offset()));
1060 testptr(mcs, mcs);
1061 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1062 bind(has_counters);
1063 }
1064
1065
1066 // Lock object
1067 //
1068 // Args:
1069 // rdx, c_rarg1: BasicObjectLock to be used for locking
1070 //
1071 // Kills:
1072 // rax, rbx
1073 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1074 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
1075
1076 Label done, slow_case;
1077
1078 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1079 const Register tmp_reg = rbx;
1080 const Register obj_reg = c_rarg3; // Will contain the oop
1081
1082 // Load object pointer into obj_reg
1083 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1084
1085 lightweight_lock(lock_reg, obj_reg, swap_reg, tmp_reg, slow_case);
1086 jmp(done);
1087
1088 bind(slow_case);
1089
1090 // Call the runtime routine for slow case
1091 call_VM_preemptable(noreg,
1092 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1093 lock_reg);
1094 bind(done);
1095 }
1096
1097
1098 // Unlocks an object. Used in monitorexit bytecode and
1099 // remove_activation. Throws an IllegalMonitorException if object is
1100 // not locked by current thread.
1101 //
1102 // Args:
1103 // rdx, c_rarg1: BasicObjectLock for lock
1104 //
1105 // Kills:
1106 // rax
1107 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1108 // rscratch1 (scratch reg)
1109 // rax, rbx, rcx, rdx
1110 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1111 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
1112
1113 Label done, slow_case;
1114
1115 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1116 const Register header_reg = c_rarg2; // Will contain the old oopMark
1117 const Register obj_reg = c_rarg3; // Will contain the oop
1118
1119 save_bcp(); // Save in case of exception
1120
1121 // Load oop into obj_reg(%c_rarg3)
1122 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1123
1124 // Free entry
1125 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1126
1127 lightweight_unlock(obj_reg, swap_reg, header_reg, slow_case);
1128 jmp(done);
1129
1130 bind(slow_case);
1131 // Call the runtime routine for slow case.
1132 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1133 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1134
1135 bind(done);
1136
1137 restore_bcp();
1138 }
1139
1140 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1141 Label& zero_continue) {
1142 assert(ProfileInterpreter, "must be profiling interpreter");
1143 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1144 testptr(mdp, mdp);
1145 jcc(Assembler::zero, zero_continue);
1146 }
1147
1148
1149 // Set the method data pointer for the current bcp.
1150 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1151 assert(ProfileInterpreter, "must be profiling interpreter");
1152 Label set_mdp;
1153 push(rax);
1154 push(rbx);
1155
1156 get_method(rbx);
1157 // Test MDO to avoid the call if it is null.
1158 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1159 testptr(rax, rax);
1160 jcc(Assembler::zero, set_mdp);
1161 // rbx: method
1162 // _bcp_register: bcp
1163 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1164 // rax: mdi
1165 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1166 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1167 addptr(rbx, in_bytes(MethodData::data_offset()));
1168 addptr(rax, rbx);
1169 bind(set_mdp);
1170 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1171 pop(rbx);
1172 pop(rax);
1173 }
1174
1175 void InterpreterMacroAssembler::verify_method_data_pointer() {
1176 assert(ProfileInterpreter, "must be profiling interpreter");
1177 #ifdef ASSERT
1178 Label verify_continue;
1179 push(rax);
1180 push(rbx);
1181 Register arg3_reg = c_rarg3;
1182 Register arg2_reg = c_rarg2;
1183 push(arg3_reg);
1184 push(arg2_reg);
1185 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1186 get_method(rbx);
1187
1188 // If the mdp is valid, it will point to a DataLayout header which is
1189 // consistent with the bcp. The converse is highly probable also.
1190 load_unsigned_short(arg2_reg,
1191 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1192 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1193 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1194 cmpptr(arg2_reg, _bcp_register);
1195 jcc(Assembler::equal, verify_continue);
1196 // rbx: method
1197 // _bcp_register: bcp
1198 // c_rarg3: mdp
1199 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1200 rbx, _bcp_register, arg3_reg);
1201 bind(verify_continue);
1202 pop(arg2_reg);
1203 pop(arg3_reg);
1204 pop(rbx);
1205 pop(rax);
1206 #endif // ASSERT
1207 }
1208
1209
1210 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1211 int constant,
1212 Register value) {
1213 assert(ProfileInterpreter, "must be profiling interpreter");
1214 Address data(mdp_in, constant);
1215 movptr(data, value);
1216 }
1217
1218
1219 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1220 int constant) {
1221 assert(ProfileInterpreter, "must be profiling interpreter");
1222 Address data(mdp_in, constant);
1223 addptr(data, DataLayout::counter_increment);
1224 }
1225
1226
1227 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1228 Register index,
1229 int constant) {
1230 assert(ProfileInterpreter, "must be profiling interpreter");
1231 Address data(mdp_in, index, Address::times_1, constant);
1232 addptr(data, DataLayout::counter_increment);
1233 }
1234
1235 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1236 int flag_byte_constant) {
1237 assert(ProfileInterpreter, "must be profiling interpreter");
1238 int header_offset = in_bytes(DataLayout::flags_offset());
1239 int header_bits = flag_byte_constant;
1240 // Set the flag
1241 orb(Address(mdp_in, header_offset), header_bits);
1242 }
1243
1244
1245
1246 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1247 int offset,
1248 Register value,
1249 Register test_value_out,
1250 Label& not_equal_continue) {
1251 assert(ProfileInterpreter, "must be profiling interpreter");
1252 if (test_value_out == noreg) {
1253 cmpptr(value, Address(mdp_in, offset));
1254 } else {
1255 // Put the test value into a register, so caller can use it:
1256 movptr(test_value_out, Address(mdp_in, offset));
1257 cmpptr(test_value_out, value);
1258 }
1259 jcc(Assembler::notEqual, not_equal_continue);
1260 }
1261
1262
1263 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1264 int offset_of_disp) {
1265 assert(ProfileInterpreter, "must be profiling interpreter");
1266 Address disp_address(mdp_in, offset_of_disp);
1267 addptr(mdp_in, disp_address);
1268 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1269 }
1270
1271
1272 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1273 Register reg,
1274 int offset_of_disp) {
1275 assert(ProfileInterpreter, "must be profiling interpreter");
1276 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1277 addptr(mdp_in, disp_address);
1278 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1279 }
1280
1281
1282 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1283 int constant) {
1284 assert(ProfileInterpreter, "must be profiling interpreter");
1285 addptr(mdp_in, constant);
1286 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1287 }
1288
1289
1290 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1291 assert(ProfileInterpreter, "must be profiling interpreter");
1292 push(return_bci); // save/restore across call_VM
1293 call_VM(noreg,
1294 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1295 return_bci);
1296 pop(return_bci);
1297 }
1298
1299
1300 void InterpreterMacroAssembler::profile_taken_branch(Register mdp) {
1301 if (ProfileInterpreter) {
1302 Label profile_continue;
1303
1304 // If no method data exists, go to profile_continue.
1305 test_method_data_pointer(mdp, profile_continue);
1306
1307 // We are taking a branch. Increment the taken count.
1308 increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1309
1310 // The method data pointer needs to be updated to reflect the new target.
1311 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1312 bind(profile_continue);
1313 }
1314 }
1315
1316
1317 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1318 if (ProfileInterpreter) {
1319 Label profile_continue;
1320
1321 // If no method data exists, go to profile_continue.
1322 test_method_data_pointer(mdp, profile_continue);
1323
1324 // We are not taking a branch. Increment the not taken count.
1325 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1326
1327 // The method data pointer needs to be updated to correspond to
1328 // the next bytecode
1329 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1330 bind(profile_continue);
1331 }
1332 }
1333
1334 void InterpreterMacroAssembler::profile_call(Register mdp) {
1335 if (ProfileInterpreter) {
1336 Label profile_continue;
1337
1338 // If no method data exists, go to profile_continue.
1339 test_method_data_pointer(mdp, profile_continue);
1340
1341 // We are making a call. Increment the count.
1342 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1343
1344 // The method data pointer needs to be updated to reflect the new target.
1345 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1346 bind(profile_continue);
1347 }
1348 }
1349
1350
1351 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1352 if (ProfileInterpreter) {
1353 Label profile_continue;
1354
1355 // If no method data exists, go to profile_continue.
1356 test_method_data_pointer(mdp, profile_continue);
1357
1358 // We are making a call. Increment the count.
1359 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1360
1361 // The method data pointer needs to be updated to reflect the new target.
1362 update_mdp_by_constant(mdp,
1363 in_bytes(VirtualCallData::
1364 virtual_call_data_size()));
1365 bind(profile_continue);
1366 }
1367 }
1368
1369
1370 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1371 Register mdp,
1372 Register reg2,
1373 bool receiver_can_be_null) {
1374 if (ProfileInterpreter) {
1375 Label profile_continue;
1376
1377 // If no method data exists, go to profile_continue.
1378 test_method_data_pointer(mdp, profile_continue);
1379
1380 Label skip_receiver_profile;
1381 if (receiver_can_be_null) {
1382 Label not_null;
1383 testptr(receiver, receiver);
1384 jccb(Assembler::notZero, not_null);
1385 // We are making a call. Increment the count for null receiver.
1386 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1387 jmp(skip_receiver_profile);
1388 bind(not_null);
1389 }
1390
1391 // Record the receiver type.
1392 record_klass_in_profile(receiver, mdp, reg2, true);
1393 bind(skip_receiver_profile);
1394
1395 // The method data pointer needs to be updated to reflect the new target.
1396 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1397 bind(profile_continue);
1398 }
1399 }
1400
1401 // This routine creates a state machine for updating the multi-row
1402 // type profile at a virtual call site (or other type-sensitive bytecode).
1403 // The machine visits each row (of receiver/count) until the receiver type
1404 // is found, or until it runs out of rows. At the same time, it remembers
1405 // the location of the first empty row. (An empty row records null for its
1406 // receiver, and can be allocated for a newly-observed receiver type.)
1407 // Because there are two degrees of freedom in the state, a simple linear
1408 // search will not work; it must be a decision tree. Hence this helper
1409 // function is recursive, to generate the required tree structured code.
1410 // It's the interpreter, so we are trading off code space for speed.
1411 // See below for example code.
1412 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1413 Register receiver, Register mdp,
1414 Register reg2, int start_row,
1415 Label& done, bool is_virtual_call) {
1416 if (TypeProfileWidth == 0) {
1417 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1418 } else {
1419 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1420 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1421 }
1422 }
1423
1424 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1425 Label& done, int total_rows,
1426 OffsetFunction item_offset_fn,
1427 OffsetFunction item_count_offset_fn) {
1428 int last_row = total_rows - 1;
1429 assert(start_row <= last_row, "must be work left to do");
1430 // Test this row for both the item and for null.
1431 // Take any of three different outcomes:
1432 // 1. found item => increment count and goto done
1433 // 2. found null => keep looking for case 1, maybe allocate this cell
1434 // 3. found something else => keep looking for cases 1 and 2
1435 // Case 3 is handled by a recursive call.
1436 for (int row = start_row; row <= last_row; row++) {
1437 Label next_test;
1438 bool test_for_null_also = (row == start_row);
1439
1440 // See if the item is item[n].
1441 int item_offset = in_bytes(item_offset_fn(row));
1442 test_mdp_data_at(mdp, item_offset, item,
1443 (test_for_null_also ? reg2 : noreg),
1444 next_test);
1445 // (Reg2 now contains the item from the CallData.)
1446
1447 // The item is item[n]. Increment count[n].
1448 int count_offset = in_bytes(item_count_offset_fn(row));
1449 increment_mdp_data_at(mdp, count_offset);
1450 jmp(done);
1451 bind(next_test);
1452
1453 if (test_for_null_also) {
1454 // Failed the equality check on item[n]... Test for null.
1455 testptr(reg2, reg2);
1456 if (start_row == last_row) {
1457 // The only thing left to do is handle the null case.
1458 Label found_null;
1459 jccb(Assembler::zero, found_null);
1460 // Item did not match any saved item and there is no empty row for it.
1461 // Increment total counter to indicate polymorphic case.
1462 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1463 jmp(done);
1464 bind(found_null);
1465 break;
1466 }
1467 Label found_null;
1468 // Since null is rare, make it be the branch-taken case.
1469 jcc(Assembler::zero, found_null);
1470
1471 // Put all the "Case 3" tests here.
1472 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1473 item_offset_fn, item_count_offset_fn);
1474
1475 // Found a null. Keep searching for a matching item,
1476 // but remember that this is an empty (unused) slot.
1477 bind(found_null);
1478 }
1479 }
1480
1481 // In the fall-through case, we found no matching item, but we
1482 // observed the item[start_row] is null.
1483
1484 // Fill in the item field and increment the count.
1485 int item_offset = in_bytes(item_offset_fn(start_row));
1486 set_mdp_data_at(mdp, item_offset, item);
1487 int count_offset = in_bytes(item_count_offset_fn(start_row));
1488 movl(reg2, DataLayout::counter_increment);
1489 set_mdp_data_at(mdp, count_offset, reg2);
1490 if (start_row > 0) {
1491 jmp(done);
1492 }
1493 }
1494
1495 // Example state machine code for three profile rows:
1496 // // main copy of decision tree, rooted at row[1]
1497 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1498 // if (row[0].rec != nullptr) {
1499 // // inner copy of decision tree, rooted at row[1]
1500 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1501 // if (row[1].rec != nullptr) {
1502 // // degenerate decision tree, rooted at row[2]
1503 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1504 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1505 // row[2].init(rec); goto done;
1506 // } else {
1507 // // remember row[1] is empty
1508 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1509 // row[1].init(rec); goto done;
1510 // }
1511 // } else {
1512 // // remember row[0] is empty
1513 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1514 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1515 // row[0].init(rec); goto done;
1516 // }
1517 // done:
1518
1519 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1520 Register mdp, Register reg2,
1521 bool is_virtual_call) {
1522 assert(ProfileInterpreter, "must be profiling");
1523 Label done;
1524
1525 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1526
1527 bind (done);
1528 }
1529
1530 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1531 Register mdp) {
1532 if (ProfileInterpreter) {
1533 Label profile_continue;
1534 uint row;
1535
1536 // If no method data exists, go to profile_continue.
1537 test_method_data_pointer(mdp, profile_continue);
1538
1539 // Update the total ret count.
1540 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1541
1542 for (row = 0; row < RetData::row_limit(); row++) {
1543 Label next_test;
1544
1545 // See if return_bci is equal to bci[n]:
1546 test_mdp_data_at(mdp,
1547 in_bytes(RetData::bci_offset(row)),
1548 return_bci, noreg,
1549 next_test);
1550
1551 // return_bci is equal to bci[n]. Increment the count.
1552 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1553
1554 // The method data pointer needs to be updated to reflect the new target.
1555 update_mdp_by_offset(mdp,
1556 in_bytes(RetData::bci_displacement_offset(row)));
1557 jmp(profile_continue);
1558 bind(next_test);
1559 }
1560
1561 update_mdp_for_ret(return_bci);
1562
1563 bind(profile_continue);
1564 }
1565 }
1566
1567
1568 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1569 if (ProfileInterpreter) {
1570 Label profile_continue;
1571
1572 // If no method data exists, go to profile_continue.
1573 test_method_data_pointer(mdp, profile_continue);
1574
1575 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1576
1577 // The method data pointer needs to be updated.
1578 int mdp_delta = in_bytes(BitData::bit_data_size());
1579 if (TypeProfileCasts) {
1580 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1581 }
1582 update_mdp_by_constant(mdp, mdp_delta);
1583
1584 bind(profile_continue);
1585 }
1586 }
1587
1588
1589 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1590 if (ProfileInterpreter) {
1591 Label profile_continue;
1592
1593 // If no method data exists, go to profile_continue.
1594 test_method_data_pointer(mdp, profile_continue);
1595
1596 // The method data pointer needs to be updated.
1597 int mdp_delta = in_bytes(BitData::bit_data_size());
1598 if (TypeProfileCasts) {
1599 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1600
1601 // Record the object type.
1602 record_klass_in_profile(klass, mdp, reg2, false);
1603 }
1604 update_mdp_by_constant(mdp, mdp_delta);
1605
1606 bind(profile_continue);
1607 }
1608 }
1609
1610
1611 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1612 if (ProfileInterpreter) {
1613 Label profile_continue;
1614
1615 // If no method data exists, go to profile_continue.
1616 test_method_data_pointer(mdp, profile_continue);
1617
1618 // Update the default case count
1619 increment_mdp_data_at(mdp,
1620 in_bytes(MultiBranchData::default_count_offset()));
1621
1622 // The method data pointer needs to be updated.
1623 update_mdp_by_offset(mdp,
1624 in_bytes(MultiBranchData::
1625 default_displacement_offset()));
1626
1627 bind(profile_continue);
1628 }
1629 }
1630
1631
1632 void InterpreterMacroAssembler::profile_switch_case(Register index,
1633 Register mdp,
1634 Register reg2) {
1635 if (ProfileInterpreter) {
1636 Label profile_continue;
1637
1638 // If no method data exists, go to profile_continue.
1639 test_method_data_pointer(mdp, profile_continue);
1640
1641 // Build the base (index * per_case_size_in_bytes()) +
1642 // case_array_offset_in_bytes()
1643 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1644 imulptr(index, reg2); // XXX l ?
1645 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1646
1647 // Update the case count
1648 increment_mdp_data_at(mdp,
1649 index,
1650 in_bytes(MultiBranchData::relative_count_offset()));
1651
1652 // The method data pointer needs to be updated.
1653 update_mdp_by_offset(mdp,
1654 index,
1655 in_bytes(MultiBranchData::
1656 relative_displacement_offset()));
1657
1658 bind(profile_continue);
1659 }
1660 }
1661
1662
1663
1664 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1665 if (state == atos) {
1666 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1667 }
1668 }
1669
1670
1671 // Jump if ((*counter_addr += increment) & mask) == 0
1672 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1673 Register scratch, Label* where) {
1674 // This update is actually not atomic and can lose a number of updates
1675 // under heavy contention, but the alternative of using the (contended)
1676 // atomic update here penalizes profiling paths too much.
1677 movl(scratch, counter_addr);
1678 incrementl(scratch, InvocationCounter::count_increment);
1679 movl(counter_addr, scratch);
1680 andl(scratch, mask);
1681 if (where != nullptr) {
1682 jcc(Assembler::zero, *where);
1683 }
1684 }
1685
1686 void InterpreterMacroAssembler::notify_method_entry() {
1687 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1688 // track stack depth. If it is possible to enter interp_only_mode we add
1689 // the code to check if the event should be sent.
1690 Register rthread = r15_thread;
1691 Register rarg = c_rarg1;
1692 if (JvmtiExport::can_post_interpreter_events()) {
1693 Label L;
1694 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1695 testl(rdx, rdx);
1696 jcc(Assembler::zero, L);
1697 call_VM(noreg, CAST_FROM_FN_PTR(address,
1698 InterpreterRuntime::post_method_entry));
1699 bind(L);
1700 }
1701
1702 if (DTraceMethodProbes) {
1703 get_method(rarg);
1704 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1705 rthread, rarg);
1706 }
1707
1708 // RedefineClasses() tracing support for obsolete method entry
1709 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1710 get_method(rarg);
1711 call_VM_leaf(
1712 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1713 rthread, rarg);
1714 }
1715 }
1716
1717
1718 void InterpreterMacroAssembler::notify_method_exit(
1719 TosState state, NotifyMethodExitMode mode) {
1720 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1721 // track stack depth. If it is possible to enter interp_only_mode we add
1722 // the code to check if the event should be sent.
1723 Register rthread = r15_thread;
1724 Register rarg = c_rarg1;
1725 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1726 Label L;
1727 // Note: frame::interpreter_frame_result has a dependency on how the
1728 // method result is saved across the call to post_method_exit. If this
1729 // is changed then the interpreter_frame_result implementation will
1730 // need to be updated too.
1731
1732 // template interpreter will leave the result on the top of the stack.
1733 push(state);
1734 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1735 testl(rdx, rdx);
1736 jcc(Assembler::zero, L);
1737 call_VM(noreg,
1738 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1739 bind(L);
1740 pop(state);
1741 }
1742
1743 if (DTraceMethodProbes) {
1744 push(state);
1745 get_method(rarg);
1746 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1747 rthread, rarg);
1748 pop(state);
1749 }
1750 }
1751
1752 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1753 // Get index out of bytecode pointer
1754 get_cache_index_at_bcp(index, 1, sizeof(u4));
1755 // Get address of invokedynamic array
1756 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
1757 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1758 if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
1759 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
1760 } else {
1761 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1762 }
1763 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
1764 }
1765
1766 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1767 // Get index out of bytecode pointer
1768 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
1769 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1770
1771 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
1772 // Take shortcut if the size is a power of 2
1773 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1774 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1775 } else {
1776 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1777 }
1778 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
1779 }
1780
1781 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1782 // Get index out of bytecode pointer
1783 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
1784 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1785
1786 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
1787 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1788 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
1789 }